This project was intended to forecast the bin volume for next 7 days based on history pattern observed and also provided an optimized collection process to save resources and get the optimal value.
The Objective:
- History Data Generation.
- Real Time Data Generation.
- LSTM Multi-step Multivariate Forecast Model Training.
- Forecast Test Data Generation.
- LSTM prediction script for 7 days forecast of bin fill level.
- Route Optimisation.
A history dataset has been generated using real life scenarios to train the model and identify the pattern while forecasting.
A similar script as above has been used to generate data on a daily basis using schedulers to automaticaly generate everyday data and store it in their respective db's.
Based on the data generated, a LSTM model with multiple layers has been trained for forecasting the percentage of bin fill level for every hour, for the next 7 days from now.
Custom funtcions has been written to generate data based on time(hourly) for the next 7 days.
On the generated data the prediction script was made, which is used to give the predictions as requested.
Based on the predicted bins for a particular day, an optimised collection route is generated, which shows which bins to be collected at first to get the shortest distance travelled for completing the collection process.